The Analysis of Social Networks

The Analysis of Social Networks Ronald L. Breiger Department of Sociology University of Arizona Tucson, AZ 85721-0027 [email protected] This document contains excerpts from a close-to-final draft of a chapter now published as R.L. Breiger. 2004. “The Analysis of Social Networks,” pp. 505-526 in Handbook of Data Analysis, edited by Melissa Hardy and Alan Bryman. London: SAGE Publications, 2004. Ronald Breiger The Analysis of Social Networks Handbook of Data Analysis, edited by Melissa hardy and Alan Bryman. Sage Publications, 2004, pp. 505-526. Outline of chapter (click on a section to read it) From Metaphor to Data Analysis Qualitative or Quantitative? Network Basics Types of Equivalence Statistical Models for Network Structure Culture and Cognition Appendix: Computer Programs, Further Study Acknowledgement Note References The Analysis of Social Networks Ronald L. Breiger Study of social relationships among actors—whether individual human beings or animals of other species, small groups or economic organizations, occupations or social classes, nations or world military alliances—is fundamental to the social sciences. Social network analysis may be defined as the disciplined inquiry into the patterning of relations among social actors, as well as the patterning of relationships among actors at different levels of analysis (such as persons and groups). Following an introduction to data analysis issues in social networks research and to the basic forms of network representation, three broad topics are treated under this chapter’s main headings: types of equivalence, statistical models (emphasizing a new class of logistic regression models for networks), and culture and cognition. Each section emphasizes data- analytic strategies used in exemplary research studies of social networks. An appendix to the chapter briefly treats computer programs and related issues. From Metaphor to Data Analysis Network metaphors have long had great intuitive appeal for social thinkers and social scientists. Marx ( [1857] 1956: 96) held that “society is not merely an aggregate of individuals; it is the sum of the relations in which these individuals stand to one another.” Durkheim, in his Latin thesis, traced his interest in social morphology to that of the eighteenth-century thinker Montesqueiu, who had identified various types of society, such as monarchy, aristocracy, and republic “not on the basis of division of labor or the nature of their social ties, but solely according to nature of their sovereign authority,” and Durkheim went on to criticize this strategy as a failure to see “that the essential is not the number of persons subject to the same authority, but the number bound by some form of relationship” (Durkheim [1892] 1965: 32, 38). Leopold von Wiese, a writer within the German “formal school” centered around Georg Simmel, asked his reader to imagine what would be seen if “the constantly flowing stream of interhuman activity” were halted in its course for just one moment, and suggested that “we will then see that it is an apparently impenetrable network of lines between men;” furthermore, “a static analysis of the sphere of the interhuman will ... consist in the dismemberment and reconstruction of this system of relations” (Wiese, [1931] 1941: 29-30). In America, Charles Horton Cooley proclaimed the necessity of a “social” or “sociological” pragmatism (Joas 1993: 23), a tradition within which not only consciousness of social relations, but self-consciousness, was theorized explicitly, and within which “a man may be regarded as the point of intersection of an indefinite number of lines representing social groups, having as many arcs passing through him as there are groups” (Cooley, [1902] 1964: 148). The English social anthropologist A. R. Radcliffe-Brown (1940) wrote that ‘direct observation does reveal to us that ... human beings are connected by a complex network of social relations. I use the term “social structure” to denote this network....’ One prominent commentator on the history of social scientific thought and on contemporary development writes that `network sociology is doing the very thing that early sociologists and anthropologists saw as crucial—the mapping of the relations that create social structures” (Turner 1991: 571). Much contemporary research over the past decades can be seen as a move from network thinking as vague metaphors to the elaboration of the network concept to the point where it can be used as an exact representation of at least some central elements of social structure (Freeman 1989, Smith-Lovin and McPherson 1993, Wellman 1988). A particularly notable move from metaphor to analytical method is the relatively recent development of highly sophisticated computer programs for producing pictorial representations of social networks. Freeman (2000) illustrates some of the newest procedures for producing web- based pictures that allow viewers to interact with the network data and thus to use visual input in exploring a variety of analytical models of their structural properties. Qualitative or Quantitative? Many of the quantitative techniques of data analysis taken up in other chapters of this Handbook may be considered to study networks of statistical relationships among variables. In contrast, because social network actors are in most studies concrete and observable (such as individual persons, groups, nations, and alliances) or collectivities of observable agents (such as occupations or classes), the relationships of interest to an analyst of social networks are usually in the first instance social or cultural, binding together and differentiating concrete entities, rather than statistical encodings. Indeed, some social network theorists and data analysts emphasize the extent to which the inductive modeling strategies of network analyses are subversive of the usual canons of statistical methods (Levine, 1999) or they portray network analysts as "eschwing ... interactional approaches such as statistical (variable) analyses ... [in order to] pursue transactional studies of patterned social relationships" (Emirbayer, 1997: 298). Much progress has in fact been made in recent decades in the statistical analysis of social networks. A number of important extensions of the general linear model have been developed in recent decades specifically to model (for example) the likelihood of a relationship existing from one actor to another, taking into account the lack of independence among social relationships and the presence of specifiable patterns of dependence. These statistical models will be reviewed in this chapter. Nonetheless, it is useful to keep in mind the very considerable extent to which social network analysis as a strategy of empirical research is indeed difficult to contain within the conventionally established headings of statistical data analysis, to the point of suggesting a highly distinctive research orientation within the social sciences. Even the effort to maintain distinctions between quantitative and qualitative forms of data analysis is challenged by the progress that has been made in the analysis of social networks. Despite the development of important statistical models for social networks, therefore, as one strand of recent research deserving exposition, it will be useful to emphasize other strands that portray network analysis as a form of data analysis moving in directions quite different from statistical modeling. Even with respect to mathematical models, some of the most important progress in network analysis is more likely to be treated in texts on applied abstract algebra (Kim and Roush, 1983; Lidl and Pilz, 1998) than in statistics texts, owing to progress that has been made in visualizing and modeling complex structures as distinct from estimating relevant quantities. A related point is that samples of independent actors or relationships are only rarely the focus of network analysis. It has been famously suggested (Barton, 1968) that, just as a physiologist would be misled by sampling every hundredth cell of a dissected laboratory animal in an effort to understand its bodily structure and functioning, so a scientist interested in social structure and behavior would be misled by reliance on random samples wrenched out of their embedded interactional context; at a minimum, highly innovative sampling theory and methods need to be developed afresh (as they are for example in McPherson, 1982). In contrast to random samples, full data on the presence or absence of social relations among all the members of a bounded population is often required, and network analysts have formulated the problem of boundary specification (Laumann et al., 1983; Marsden, 1990) in an effort to gain analytical leverage on this requirement. Furthermore, whether in laboratory experiments furthering exchange theory (Molm and Cook, 1995) or in broad observations on new forms of "recombinant property" taking shape in post-Communist Eastern Europe (Stark, 1996), the innovations in data analysis of social networks are very often substantive rather than statistical in nature. Finally, the very distinction between “quantitative” and “qualitative” approaches to data analysis is called into question by network analysis, in ways that go beyond the distinction between (quantitative) statistics and (qualitative) algebra. Typically a network analysis is a case study (Ragin and Becker, 1992) situated with explicit temporal and spatial reference (Abbott, 1999: 193-226), and important contributions to data analysis have

Load more